feat: add streaming learning assistant and knowledge base scaffolding

app/agents/learning_assistant_agent.py

@@ -0,0 +1,58 @@
+from collections.abc import AsyncIterator
+
+from app.agents.llm_adapter import LLMResponse, LLMStreamChunk, OpenAICompatibleLLMClient
+from app.core.config import settings
+from app.schemas.learning_assistant import LearningAssistantSource
+
+
+class LearningAssistantAgent:
+    """AI学习助手 Agent：根据 RAG 来源生成带循证出处的医学学习回答。"""
+
+    def __init__(self, llm_client: OpenAICompatibleLLMClient | None = None) -> None:
+        self.llm_client = llm_client or OpenAICompatibleLLMClient()
+
+    async def answer(self, question: str, sources: list[LearningAssistantSource]) -> LLMResponse:
+        """非流式回答：把问题和检索来源拼接后调用快速模型生成标准回答。"""
+        return await self.llm_client.chat(
+            self._messages(question, sources),
+            model=settings.llm_fast_model,
+            thinking_enabled=settings.llm_fast_thinking_enabled,
+            max_tokens=1200,
+        )
+
+    async def stream_answer(self, question: str, sources: list[LearningAssistantSource]) -> AsyncIterator[LLMStreamChunk]:
+        """流式回答：输出 AI 学习助手增量文本，前端可直接渲染。"""
+        async for chunk in self.llm_client.stream_chat(
+            self._messages(question, sources),
+            model=settings.llm_fast_model,
+            thinking_enabled=settings.llm_fast_thinking_enabled,
+            max_tokens=1200,
+        ):
+            yield chunk
+
+    def _messages(self, question: str, sources: list[LearningAssistantSource]) -> list[dict]:
+        """提示词拼接：命中知识库时必须引用来源，未命中时必须声明未找到参考。"""
+        if sources:
+            context = "\n\n".join(
+                (
+                    f"[来源{index}] 文档：{source.document_title or source.file_name}；"
+                    f"页码：{source.page_start}-{source.page_end}；chunk_uid：{source.chunk_uid}\n"
+                    f"{source.quote}"
+                )
+                for index, source in enumerate(sources, start=1)
+            )
+            system = (
+                "你是医学学习助手，只用于医学教育学习，不替代临床诊疗。"
+                "请优先依据给定知识库片段回答，回答要清晰、准确、分点。"
+                "每个关键结论后标注对应来源编号，例如【来源1】。"
+                "不得编造不存在的PDF、页码或指南来源。"
+            )
+            user = f"用户问题：{question}\n\n可用知识库片段：\n{context}\n\n请给出带来源的学习回答。"
+        else:
+            system = (
+                "你是医学学习助手，只用于医学教育学习，不替代临床诊疗。"
+                "当前没有检索到机构知识库参考，回答开头必须写：未检索到本机构知识库参考，以下为大模型通用学习回答。"
+                "不得伪造PDF来源、页码或指南名称。"
+            )
+            user = f"用户问题：{question}\n\n请给出通用学习回答，并提醒用户以课程教材和临床规范为准。"
+        return [{"role": "system", "content": system}, {"role": "user", "content": user}]